Privacy
Pilot draft. This notice will be reviewed before general release.
Child records stay on the desktop
BrightPath stores child profiles, photos, learning material, questions, answers, progress, drafts, analysis and approval history on your own computer. BrightPath Cloud does not persist child lists, child identifiers, learning content or content fingerprints.
After installation and local model setup, BrightPath can work offline. Use downloaded models in Ollama or LM Studio on your computer or home network. ChatGPT and other external AI providers require Internet access.
What happens when you allow ChatGPT access
Installing the plugin is not permission to read child data. A parent signs in on the desktop, selects children and permissions, and explicitly accepts the sharing explanation. The Cloud consent page starts with all connections unchecked. You can connect for generic help without selecting a family connection.
For a permitted tool request, the desktop checks current permissions and returns the requested information through Cloud to ChatGPT. Cloud processes these requests and responses in memory while relaying them; it does not save child records in its database. No child content is included in application request logs. Review and approval records are saved on the desktop.
The analysis permission shares progress and skills so ChatGPT can discuss subject-specific evidence and suggest next steps. Answers and writing need a separate permission. The desktop also offers subject analysis without sending it to ChatGPT.
OpenAI processes information received by ChatGPT under its own terms and data controls. The same applies to another AI provider you choose. This notice makes no zero-retention promise about those providers. Pasting child information into a chat also shares it with that service, independently of BrightPath’s tool permissions.
Account and service metadata in Cloud
- During the controlled pilot, your real name, self-reported country/region, introduction, email verification and contact-confirmation status are stored with your account so authorized reviewers can contact you and decide participation. These application details are removed when you erase or delete your account. Do not submit children’s details or identity documents.
- Email, display name, password hash and access decision: kept with your account until deletion. Unverified accounts expire after 7 days.
- Computer identities, connection permissions and expiry dates: retained while connected, then normally removed 30 days after revocation, expiry or disconnection. Child selections remain on the desktop.
- Sign-in, verification, reset, installer and OAuth token hashes: removed after expiry or consumption as applicable. Revoked ChatGPT links are removed after 30 days.
- Operational records contain opaque operation identifiers, action names, times, status and error codes, without child identifiers or content fingerprints. They are removed 7 days after their last update.
- If you use Cloud AI models, account-level request, token and cost totals are kept for the current and previous month. No prompts or answers are stored in these totals.
Expiry cleanup normally runs hourly. These periods describe the active application database; infrastructure logs and backup handling must also be covered in the completed operator privacy notice before general release.
Public website statistics
The public BrightPath website keeps aggregate daily counts for public page views and installer downloads. Counts are grouped only by page, referring website host and broad device category. BrightPath does not store visitor IP addresses, cookies, browser fingerprints, account identifiers, installation identifiers or child information for these statistics. Public-site statistics are collected by Cloud only and are not sent by or to the BrightPath desktop application.
Optional technical support
Support is separate from permission to access child records. If you prepare a support draft, Cloud stores the technical fields you provide, your account ownership and notification choice. Submitting requires review of the exact version. Authorized support operators can read submitted technical reports and record findings and replies. Do not include child records, contact details, health information, passwords, logs or attachments. Automatic checks cannot detect all private information.
Unsubmitted Cloud drafts expire after seven days by default. Closed or resolved tickets are removed after ninety days by default; submission receipts may remain for up to thirty days to handle retries. Account deletion removes the associated Cloud support records. You can change resolution-email preferences on each ticket and remove local drafts in Desktop.
Investigation projects remain ordinary local Ambriya projects on the processing installation. Cloud stores project identifiers and work metadata, not project files or transcripts. Existing Ambriya project mobility is used to reopen a project later or elsewhere.
Account email and hosting
Account verification, password reset and access-decision messages are sent using an email delivery provider. Cloud hosting and email delivery process the technical information needed to provide those services. Being developed in Europe does not itself mean Cloud is hosted in Europe.
Your choices
Revoke a desktop connection to stop future tool access, unlink ChatGPT from the Cloud dashboard, or delete your Cloud account there. Account deletion does not delete your local learning records. Revocation cannot retract information already received by ChatGPT or another provider.
For privacy questions, contact birhanu.discourse@gmail.com.
Safety checks
BrightPath reviews requests and learning content using the family's selected safety model. Local models keep these checks on the computer; choosing an external provider sends the reviewed content to that provider. Cloud model requests also receive safety checks with the selected Cloud provider. These checks do not create a Cloud copy of child records. Blocked generated chat answers are replaced by a refusal; the desktop keeps the safety flag. BrightPath does not diagnose children. Please seek qualified medical advice for medical or psychological concerns.
Safety models can make mistakes. BrightPath blocks delivery when a required check fails or gives an unclear result, and parent approval cannot bypass a refusal. BrightPath cannot control ChatGPT replies outside its tools.